spanning trees, forests and limit shapes
- R. Kenyon (Brown University)
spanning trees, forests and limit shapes R. Kenyon (Brown - - PowerPoint PPT Presentation
spanning trees, forests and limit shapes R. Kenyon (Brown University) UST on Z 2 Prob(degree = 2) = 1 trunk properties (K, Wilson): 2 2 1) k Prob( ) = ( { k 3 Given a graph G : 4 2 1 Let d : R V R E be the incidence matrix:
UST on Z2
trunk properties (K, Wilson): Prob(degree = 2) = 1 2 Prob( ) = ( √ 2 − 1)k
k
Let d : RV → RE be the incidence matrix:
1 2 3 4
1 −1 1 −1 1 −1 1 −1 1 −1 Define ∆ = d∗Cd, where C is a diagonal matrix of conductances. Thm (Kirchhoff 1865) Given a graph G: ∆f(v) = X
w∼v
cvw(f(v) − f(w))
remove a row and column from ∆
det ∆0 = X
Y
e
ce
Z2 What do the roots of P(z, w) = ˆ ∆ tell us about ∆?
∆f(v) = 4f(v) − f(v + 1) − f(v + i) − f(v − 1) − f(v − i)
∆ is a “convolution” operator; its Fourier transform is multiplication by P(z, w):
Example G =
P(z, w) = 4 − z − 1 z − w − 1 w. Q.
G(x, y) = − 1 4π2 Z zxwy − 1 4 − z − 1/z − w − 1/w dz dw The Green’s function T2 For large x, y, the only relevant part of P = 0 is near (1, 1). (potential kernel)
2 2 2 2 2 2 2 2
Periodic conductances: ˆ ∆ = ✓ 5 − w − 1/w −2 − 1/z −2 − z 5 − w − 1/w ◆ P(z, w) = det ˆ ∆ = w2 + 1 w2 − 10w − 10 w − 2z − 2 z + 22 What about a slightly different setting?
2 4
2 4
P = 0 has topology away from (z, w) = (1, 1): log |z| log |w| Combinatorially, what is the meaning of the coefficients of P?
Q.
indexed by points (z, w) on P(z, w) = 0. UST ↔ (1, 1) Other points correspond to measures on “essential spanning forests”
sample configuration from another measure (triangular lattice)
µ1 µ2
µ3
µ4
On a strip graph,
“Cube groves” of Carroll/Speyer were discovered in the study the cube recurrence. (2004)
N1 : P(z, w) counts cycle-rooted spanning forests (CRSFs) (subgraphs in which each component is a tree plus one edge)
P(z, w) = w2 + 1 w2 − 10w − 10 w − 2z − 2 z + 22
Thm: where C has k cycles of homology class (i, j). P(z, w) = X
CRSFs C
Y
e
ce ! (2 − ziwj − z−iw−j)k, The set of homology classes forms a convex polygon N:
= 2(2 − z − 1 z ) + (2 − w − 1 w)2 + 6(2 − w − 1 w)
e.g.
(+. . . ) the Newton polygon of P, symmetric around (0, 0)
N10 : If we enlarge the fundamental domain (take a cover of the torus) for the 10 × 10 cover
Real points (s, t) in N1 parametrize measures µs,t on planar configurations
with fixed average slope and density of crossings:
a random sample from µ.3,.5.
(N10 = 10N1)
UST
CRSFs with average slope 1/2
Thm: The measures µs,t are determinantal (for edges). The kernel is given by Pr(e1, . . . , ek ∈ T) = det[K(ei, ej)i,j=1,...,k] *There is a matrix K such that * Note: The only dependence on s, t is in contour of integration. (Ks,t)e,f = 1 4π2 ZZ
|z|=eX,|w|=eY
K(z, w)[e],[f]zx1−x2wy1−y2 P(z, w) dz iz dw iw
10
20 500 1000
σ(s, t) = lim
n→∞
1 n2 log([zsnwtn]Pn×n(z, w)) is the growth rate of the appropriate coefficient of Pn×n(z, w): The free energy (growth rate) of µs,t
2 4
2 4
N1 :
uncritical points
the amoeba of P x = log |z| y = log |w| s, t-plane x, y-plane rσ rR For s, t an uncritical point, µs,t has exponential decay of correlations!
5 10
5 10
If the graph G has larger fundamental domain, the phase space is richer:
log|z| log|w|
Boundary connections and limit shapes
Given a region with a spanning forest connecting certain boundary vertices...
Find a uniform spanning forest with the same boundary connections...
More generally, start with a CRSF on a multiply connected domain (possibly having certain boundary connections)...and find a uniform sample with same homotopy type and connections.
Given a domain U ⊂ R2 and “unsigned 1-form” |dy| satisfying a certain Lipschitz condition: |dy(u)|u ∈ N for u ∈ S1 approximating |dy| exists, and is the unique unsigned 1-form maximizing the limit of the CRSF process |dy✏| on U ∩ ✏Z2 with local slope ZZ
U
σ(|dy|) dA.
“measured foliation”
How to sample CRSFs with given topology? ...MCMC add and remove “cubes” whose faces are decorated with spanning tree edges (or dual edges).